Mechanisms of splicing-dependent trans-synaptic adhesion by PTPδ–IL1RAPL1/IL-1RAcP for synaptic differentiation

Synapse formation is triggered through trans-synaptic interaction between pairs of pre- and postsynaptic adhesion molecules, the specificity of which depends on splice inserts known as ‘splice-insert signaling codes'. Receptor protein tyrosine phosphatase δ (PTPδ) can bidirectionally induce pre- and postsynaptic differentiation of neurons by trans-synaptically binding to interleukin-1 receptor accessory protein (IL-1RAcP) and IL-1RAcP-like-1 (IL1RAPL1) in a splicing-dependent manner. Here, we report crystal structures of PTPδ in complex with IL1RAPL1 and IL-1RAcP. The first immunoglobulin-like (Ig) domain of IL1RAPL1 directly recognizes the first splice insert, which is critical for binding to IL1RAPL1. The second splice insert functions as an adjustable linker that positions the Ig2 and Ig3 domains of PTPδ for simultaneously interacting with the Ig1 domain of IL1RAPL1 or IL-1RAcP. We further identified the IL1RAPL1-specific interaction, which appears coupled to the first-splice-insert-mediated interaction. Our results thus reveal the decoding mechanism of splice-insert signaling codes for synaptic differentiation induced by trans-synaptic adhesion between PTPδ and IL1RAPL1/IL-1RAcP.

M ammalian brains are composed of at least a few hundred billions of neurons, which are connected by synapses in a spatiotemporally organized manner to establish higher-order brain functions such as learning, memory and emotion. Synapse formation is triggered through transsynaptic interactions between selective pairs of pre-and postsynaptic adhesion molecules known as 'synaptic organizers' or 'synaptogenic proteins', many of which are associated with neurodevelopmental disorders such as intellectual disability and autism [1][2][3][4][5][6] . Recent studies have highlighted a major role of type-IIa receptor protein tyrosine phosphatases (RPTPs) as synaptic organizers in pre-and postsynaptic differentiation during neural development 5,6 . Vertebrate type-IIa RPTPs comprise LAR, PTPs and PTPd, which share common domain architecture with a single transmembrane (TM) helix flanked by a large extracellular domain (ECD) and cytoplasmic tandem PTP domains: the membrane-proximal PTP domain (D1) is catalytically active, whereas the membrane-distal PTP domain (D2) is inactive. The ECD consists of three immunogloblin-like (Ig) domains and four to eight fibronectin type-III (Fn) domains. In mammals, presynaptic type-IIa RPTPs form trans-synaptic adhesion by specifically binding to their cognate postsynaptic organizers through their ECDs for inducing synaptic differentiation of neurons. To date, five different types of postsynaptic organizers for type-IIa RPTPs have been reported: interleukin-1 receptor (IL-1R) accessory protein (IL-1RAcP) 7 , IL-1RAcP-like-1 (IL1RAPL1) 8 , neurotrophin receptor tyrosine kinase C (TrkC) 9 , netrin-G ligand-3 (NGL-3) 10,11 and Slit-and Trk-like (Slitrk) family proteins 12,13 . For example, co-culture with either IL-1RAcP-or IL1RAPL1-expressing non-neuronal HEK293T cells can stimulate presynaptic differentiation in contacting axons of cultured cortical neurons, however, the IL1RAPL1and IL-1RAcP-induced presynaptic differentiations are completely abolished and decreased by B70% in the neurons lacking PTPd, respectively 7,8 . Similarly, PTPd can induce postsynaptic differentiation of cultured cortical neurons in a manner specific to IL1RAPL1 and IL-1RAcP (more precisely, a central nervous system-restricted IL-1RAcP isoform, which differs only in the carboxy (C)-terminal from a widely distributed isoform). On the other hand, Slitrk family proteins induce presynaptic differentiation through the interaction with type-IIa RPTPs 12,13 . Among Slitrk family proteins, Slitrk3 selectively induces inhibitory synapse formation, suggesting potential roles of synapse organizers in balancing excitatory and inhibitory synapses 12 . Various combinations of trans-synaptic adhesion between pre-and postsynaptic organizers might play an important role in generating extremely diverged but highly organized synaptic connections of neurons.
Varieties of the combinations between pre-and postsynaptic organizers rely upon structural diversities of their ECDs, which can be drastically increased by multiple isoforms and splice variants. In fact, several isoforms and splice variants of type-IIa RPTPs are expressed in mouse brain, and bind selectively to the corresponding postsynaptic organizers 5,6 . Type-IIa RPTPs commonly have three Ig domains (Ig1-Ig3) in their amino (N)-terminal regions. The first splicing site is located within Ig2, whereas the second one is in the junction between Ig2 and Ig3. These two splicing sites generate variants of the type-IIa RPTPs that lack or contain short-peptide inserts termed mini-exon peptides. A recent study has shown that Slitrk family members bind to all type-IIa RPTP members, depending on the second mini-exon peptide (meB) 14 . On the other hand, IL1RAPL1 and IL-1RAcP specifically bind to PTPd but not to LAR or PTPs. Furthermore, our previous cell-surface binding assays showed that both the first mini-exon peptide (meA) and meB can alter their binding properties 7,8 . MeB comprises four residues, Glu-Leu-Arg-Glu, whereas there are three variations in meA: a three-residue peptide Glu-Ser-Ile (meA3), a six-residue peptide Gly-Gly-Thr-Pro-Ile-Arg (meA6) and their tandem combination (Glu-Ser-Ile-Gly-Gly-Thr-Pro-Ile-Arg) (meA9). Binding to IL-1RAcP is significantly enhanced by the presence of both meA and meB, independently of the meA variation. In contrast, the length and sequence of meA are critically important for binding to IL1RAPL1. Only PTPd variants containing meA9 and meA6 can bind to IL1RAPL1. The meB insertion increases binding of the meA9-containing variant, whereas the meA6containing variant absolutely requires the combination with meB. These findings indicate that information of synaptogenic postsynaptic ligands for the type-IIa RPTPs is encoded in their splice inserts. Similarly, in another presynaptic organizer neurexin, the insert in the splice site 4 regulates the specificity to postsynaptic organizers Cbln1-GluRd2 and LRRTM1/2 (refs 15,16). The 'splice-insert signaling code' concept for specific synaptic connections was proposed originally for transsynaptic adhesion between neurexin and neuroligin, but appears likely more general for synaptogenic adhesion 3,7-9,15-18 . However, structural mechanisms underlying this concept remain elusive. To elucidate the structural basis for decoding the 'splice-insert signaling code' in the type-IIa RPTPs, we determined the crystal structures of PTPd-ECD in complex with IL-1RAcP-ECD and IL1RAPL1-ECD. Together with structurebased mutational analyses by surface plasmon resonance (SPR) spectroscopy and synaptogenic assays, we reveal the structural basis of the splicing-dependent trans-synaptic adhesion by PTPd and IL1RAPL1/IL-1RAcP for synaptic differentiation.

Results
Structure of the complex between PTPd and IL1RAPL1. To elucidate the structural basis of the PTPd-IL1RAPL1 interaction for trans-synaptic adhesion, we determined the crystal structures of IL1RAPL1-ECD in complex with PTPd Ig1-Ig2 and the fulllength PTPd-ECD at 2.7 and 4.4 Å resolutions, respectively ( Fig. 1a,b, Table 1). Description of PTPd Ig1-Ig2 will be based on the 2.7-Å-complex structure, whereas that of the other regions of PTPd-ECD will be based on the 4.4-Å-complex structure. For crystallography, we used a PTPd isoform containing four Fn domains and both meA9 and meB, which is the most abundant in the developing brain and has the highest affinity for IL1RAPL1 (K d : 0.15 mM) among all the PTPd variants (Table 2). Hereafter, 'PTPd' indicates this isoform, unless otherwise noted. The Ig1-Ig2 domains of PTPd retain the binding ability for IL1RAPL1-ECD, as shown in our serial deletion analysis of PTPd ( Supplementary  Fig. 1a). PTPd-ECD binds to IL1RAPL1-ECD at a ratio of 1:1 and the asymmetric unit contains one PTPdIL1RAPL1 complex ( Fig. 1b and Supplementary Fig. 1b). IL1RAPL1-ECD consists of three Ig domains, which are arranged in an L-shape (Fig. 1a,b). PTPd-ECD exhibits an elongated shape with a length of 180 Å along the longest axis. The Ig1 and Ig2 domains of PTPd tightly interact with each other to form a compact V-shaped unit, which is almost identical to the isolated Ig1-Ig2 structures of PTPd 19 (PDB:2YD6, 2YD7) (rmsd values of 0.8-0.9 Å for 195 Ca atoms). No substantial change in the backbone structure of PTPd Ig1-Ig2 occurred upon binding to IL1RAPL1 ( Supplementary Fig. 1c). The Ig3 domain is spatially separated by the meB-containing linker and positioned apart from the Ig1-Ig2 unit. The following Fn1 and Fn2 domains are aligned in a straight line, similarly to the isolated PTPd Fn1-Fn2 structure, which we determined at 2.0 Å resolution ( Supplementary Fig. 1d). The Fn3 domain of PTPd is oriented perpendicular to the Fn1-Fn2 domains, as observed in the PTPs Ig1-Fn3 structure ( Supplementary Fig. 1e). Electron density of the Fn4 domain and subsequent C-terminal region was not visible, likely due to the disorder. The length of the PTPd-ECDIL1RAPL1-ECD complex along the longest axis is 206 Å, which is equivalent to the average length of excitatory synaptic clefts, implying that PTPd-ECD is kinked at the junction between the Fn2 and Fn3 domains in synapses.
Interactions between IL1RAPL1 Ig1 and PTPd Ig2 domains. The Ig1 domain of IL1RAPL1 interacts with both Ig2 and Ig3 domains of PTPd (Fig. 1b). The interface between the IL1RAPL Ig1 and PTPd Ig2 domains comprises both hydrophobic and hydrophilic interactions with a buried surface area of 699 Å 2 . These interactions occur on two b-strands comprising Arg181-Ser187 and Arg196-Glu202 in the Ig2 domain of PTPd, between which meA9 is inserted (Fig. 1c). These two b-strands are hereafter referred to as the IL1RAPL1/IL-1RAcP-Ig1interacting b-strands, because they also mediate the interaction with the Ig1 domain of IL-1RAcP as described later. The first eight residues of meA9 ( 188 Glu-Ser-Ile-Gly-Gly-Thr-Pro-Ile 195 ) form a loop connecting these two b-strands, whereas the last residue, Arg196, extends one of the two b-strands. Trp34 of IL1RAPL1 hydrophobically interacts with Leu153, Ala198 and Leu185 of PTPd (Fig. 1c). The affinity of the IL1RAPL1 W34A mutant for PTPd could not be measured by surface-plasmon resonance (SPR) spectroscopy, due to substantially low signals, indicating that this hydrophobic interaction is essential for binding between PTPd and IL1RAPL1. The aliphatic portion of PTPd Arg196 also participates in this hydrophobic interaction, which enables the Ig2 domain of PTPd to more tightly interact with the Ig1 domain of IL1RAPL1 (Fig. 1c) Table 3). The meA6-containing PTPd variant also contains Arg196 but has a fivefold reduced affinity for IL1RAPL1 (Table 2). Therefore, this affinity reduction of the meA6-containing variant might be due to a lack of the concomitant hydrogen bonding. The atomic-level meA-specific interactions described here illustrate how the meA insertion contributes to the affinity between PTPd and IL1RAPL1.
On the opposite side to the IL1RAPL1 Ig1-PTPd Ig2 interface, the Ig1 domain of IL1RAPL1 interacts with the Ig3 domain of PTPd with a buried surface area of 694 Å 2 . The IL1RAPL1 Ig1-PTPd Ig3 interface is stabilized by hydrophobic interactions (Fig. 1d). The side chains of IL1RAPL1 Tyr77 and PTPd Tyr273 are aligned in an antiparallel manner to form a hydrophobic core, which is surrounded by Met75, Pro88 and Phe91 of IL1RAPL1 and Pro270, Met271, Ile291 and Thr314 of PTPd. The Y273A mutation of PTPd decreased the affinity for IL1RAPL1 sixfold, whereas the M75A/Y77A/P88A/F91A quadruple mutation of IL1RAPL1 decreased the affinity for PTPd 19-fold ( Table 3). The meB insertion of PTPd is important for binding between PTPd and IL1RAPL1. The lack of meB reduces the affinity fivefold (Table 2). In the present PTPd-ECDIL1RAPL1-ECD complex, the meB insertion appears to adjust the relative spacing    and orientation between the Ig2 and Ig3 domains of PTPd (Fig. 1b). When the apo structure of PTPd Ig1-Fn2 containing meA3 and lacking meB (PTPdA3B À ) ( Supplementary Fig. 1f), which we determined at 3.5 Å resolution, is superposed onto the IL1RAPL1-ECD-bound structure of PTPd Ig1-Fn2 ( Supplementary Fig. 1g) using the Ig1-Ig2 unit of PTPd as the reference, the Ig3 domain of the apo PTPdA3B À is oriented in the opposite direction from that of the IL1RAPL1-ECD-bound PTPd and cannot interact with IL1RAPL1, even if structural flexibility of the linker between the Ig2 and Ig3 domains of PTPd is assumed (Supplementary Fig. 1g). Replacement of the native meB (Glu-Leu-Arg-Glu) by two tandem units of meB or a Gly-Ser-Ser-Gly or Gln-Leu-Glu-Gln tetrapeptide hardly affected the affinity, whereas that by three tandem units reduced the affinity threefold (  PTPd Ig1 interface is overlapped with the putative binding pocket for glycosaminoglycan chains of heparan sulfate and chondroitin sulfate proteoglycans ( Supplementary Fig. 1h), which mediate axonal growth control though the interaction with the type-IIa RPTPs 19 . This third interface is formed primarily by hydrophilic interactions (Fig. 1e): a positively charged cluster comprising Arg75, Arg95 and Arg98 of PTPd forms a hydrogen bond network with a negatively charged cluster comprising Glu291 and Asp292 of IL1RAPL1. This network is stabilized by a cation-p interaction between Phe289 of IL1RAPL1 and Arg98 of PTPd. In addition, Arg124 of PTPd forms a hydrogen bond with Glu337 of IL1RAPL1. To confirm the importance of this interface, IL1RAPL1 Asp292-PTPd Arg75 interacting pair was disrupted by Ala replacement. The D292A mutation of IL1RAPL1 and the R75A mutation of PTPd reduced the affinity between IL1RAPL1 and PTPd 15-and 19-fold, respectively (Table 3). When comparing the amino-acid sequences between IL1RAPL1 and IL-1RAcP, the Ig3 domain is the least conserved domain (23% identity) among three Ig domains. As described below, the Ig3 domain of IL-1RAcP does not interact with PTPd in the PTPdIL-1RAcP complex. Therefore, the negatively charged cluster of the IL1RAPL1 Ig3 domain and the positively charged cluster of the PTPd Ig1 domain are another determinant for selective binding between PTPd and IL1RAPL1.
Structure of the complex between PTPd and IL-1RAcP. IL-1RAcP, a member of the same IL-1R family as IL1RAPL1 (B30% sequence identity), also functions as a postsynaptic organizer for PTPd. However, the affinity of IL-1RAcP for PTPd (K d : 0.68 mM) is fivefold lower than that of IL1RAPL1 ( Table 2). The meA variation hardly affects the affinity. In addition, IL-1RAcP can bind to the PTPd variant lacking both meA and meB (K d : 2.4 mM) in contrast to IL1RAPL1. To structurally explain these differences between IL1RAPL1 and IL-1RAcP, we also determined the crystal structure of IL-1RAcP-ECD in complex with PTPd containing three Ig domains and two Fn domains (Ig1-Fn2) at 3.25 Å resolution (Fig. 2a,b, Table 1). We had confirmed that PTPd Ig1-Fn2 efficiently binds to IL-1RAcP-ECD by testing various lengths of PTPd ( Supplementary Fig. 2a). The Ig1-Ig2 unit of PTPd is insufficient for binding to IL-1RAcP.
Similarly to the PTPdIL1RAPL1 complex, the Ig1-Fn2 domains of PTPd bind to IL-1RAcP-ECD at a ratio of 1:1 (Fig. 2b). The asymmetric unit contains one PTPdIL-1RAcP complex ( Supplementary Fig. 2b). The crystal packing patterns of the PTPdIL1RAPL1 and PTPdIL-1RAcP complexes in this study are totally different from each other and not suggestive of their lateral associations in synapses. Although it can be assumed that activations of PTPd and IL1RAPL1/IL-1RAcP are mediated by a ligand-induced monomer-dimer transition by analogy with those of receptor tyrosine kinases [20][21][22][23] and Toll/IL-1-receptor families 24 , our crystallographic and multiangle light scattering analyses using the ECDs of PTPd, IL-1RAcP and IL1RAPL1 could not support this activation mechanism ( Supplementary  Fig. 1b, 2b,c). The Ig1-Fn2 domains of PTPd exhibit an elongated shape with a length of 160 Å along the longest axis. Although overall shape of the IL-1RAcP-bound PTPd Ig1-Fn2 resembles that of the IL1RAPL1-bound PTPd Ig1-Fn2, the position and/or orientation of each domain are somewhat different between them. The Ig1-Ig3 domains of IL-1RAcP are arranged in an L-shape similar to those of IL1RAPL1, except that the relative orientation between the Ig1 and Ig2-Ig3 domains is changed by 20 degree (Supplementary Fig. 2d). Correspondingly, the Ig3 domain of IL-1RAcP does not interact with PTPd in contrast to that of IL1RAPL1. Superposition between the Ig3 domains of IL1RAPL1 and IL-1RAcP shows that the binding surface of IL1RAPL1 for the Ig1 domain of PTPd is not conserved in IL-1RAcP ( Supplementary Fig. 2e).
Interactions between IL-1RAcP Ig1 and PTPd Ig2. Similarly to the PTPdIL1RAPL1 complex, interactions between the Ig1 domain of IL-1RAcP and the Ig2 domain of PTPd occur on the IL1RAPL1/IL-1RAcP-Ig1-interacting b-strands (Fig. 2c). The interface between the IL-1RAcP Ig1 and PTPd Ig2 domains comprises both hydrophobic and hydrophilic interactions with a buried surface area of 726 Å 2 . Trp27 of IL-1RAcP hydrophobically interacts with Leu153, Ala198 and Leu185 of PTPd (Fig. 2c). The W27A mutation of IL-1RAcP reduced the affinity sevenfold (Table 4). Although Arg196 of PTPd is positioned in close proximity to Asp30 of IL-1RAcP, no observed electron density of its side chain guanidino group indicates that the interaction between Arg196 of PTPd and Asp30 of IL-1RAcP is substantially weak (Supplementary Fig. 2f). Thus, the D30A mutation of IL-1RAcP and the R196A mutation of PTPd hardly affected the affinity (Table 4). Basically, meA does not extensively interact with IL-1RAcP in the PTPd Ig1-Fn2IL-1RAcP-ECD complex, consistent with the finding that the variation of meA does not affect the affinity for IL-1RAcP (Table 2). Only one meA-mediated interaction is a hydrogen bond between the main-chain N atom of PTPd Glu188 and the main-chain O atom of IL-1RAcP Phe53, which could be formed in the shorter mini-exon variants, meA6 and meA3, but not in the variant lacking meA. The complete deletion of meA shortens the IL1RAPL1/IL-1RAcP-Ig1-interacting b-strands, which might disturb and weaken the hydrophobic interaction of IL-1RAcP Trp27 with Leu153, Ala198 and Leu185 of PTPd, resulting in the threefold reduced affinity to IL-1RAcP (Table 2; compare meA9/ þ with À / þ ). These structural features may reflect that the meA insertion increases the affinity between PTPd and IL-1RAcP, independently of its variation.
The position and orientation of the PTPd Ig3 domain relative to the IL-1RAcP Ig1 domain are similar to those relative to the IL1RAPL1 Ig1 domain (Figs 1b and 2b). The interface between the IL-1RAcP Ig1 and PTPd Ig3 domains buries a surface area of 906 Å 2 . The inside of the interface is stabilized by more tightly packed hydrophobic interactions than that of the IL1RAPL1 Ig1-PTPd Ig3 interface (Figs 1d and 2d). The side chains of IL-1RAcP Tyr71 and PTPd Tyr273 are stacked in an antiparallel manner to form a hydrophobic core, which is surrounded by the side chains of IL-1RAcP Phe85, Pro82 and Ile69 and PTPd Pro270 and Ile291 and the aliphatic portion of PTPd Thr314. The side chains of IL-1RAcP Tyr71 and PTPd Tyr273 are stabilized by hydrogen bonds with the main-chain O atoms of PTPd Met271 and IL-1RAcP Trp70, respectively. Furthermore, the Ig1 domain of IL-1RAcP recognizes a negatively charged loop comprising Glu286, Asp287 and Asp288 of PTPd. Tyr58 and Arg86 of IL-1RAcP hydrogen bond with Glu286 and Asp288 of PTPd, respectively, whereas Lys94 of IL-1RAcP hydrogen bonds with both Glu286 and Asp287 of PTPd (Fig. 2e). Conformation of this negatively charged loop is supported by Met289 of PTPd. In addition, Thr67 of IL-1RAcP hydrogen bonds with Lys275 of PTPd. Mutations of the IL-1RAcP Ig1-PTPd Ig3 interface impaired the binding affinity more drastically than those of the IL1RAPL1 Ig1-PTPd Ig3 interface: the Y273A mutation of PTPd and the I69A/Y71A/P82A/F85A quadruple mutation of IL-1RAcP completely eliminated the binding ability, whereas the K94A mutation of IL-1RAcP and the E286A mutation of PTPd decreased the affinity 12-and 8-fold, respectively (Table 4). Therefore, the interface mediated by PTPd Ig3 is more critically important for binding to IL-1RAcP than for binding to IL1RAPL1, consistent with the finding that PTPd Ig3 is essential for binding to IL-1RAcP but not to IL1RAPL1 ( Supplementary  Figs 1a and 2a). Replacement of the native meB (Glu-Leu-Arg-Glu) by a Gly-Ser-Ser-Gly or Gln-Leu-Glu-Gln tetrapeptide hardly affected the affinity for IL-1RAcP, whereas that by two or three tandem units reduced the affinity two-or sixfold, respectively (Table 4). These impacts on the affinity for IL-1RAcP are similar to those for IL1RAPL1, supporting the notion that the meB insertion acts as an adjustable linker to control relative positions and orientations of the PTPd Ig2 and Ig3 domains for their simultaneous interactions with the Ig1 domain of IL-1RAcP.
Synaptogenic activity. Finally, we investigated relationships between the structures and synaptogenic activity (that is, IL1RAPL1/IL-1RAcP-induced presynaptic differentiation and PTPd-induced postsynaptic differentiation). IL1RAPL1/IL-1RAcP mutations were examined by fibroblast-neuron mixed culture assays (Fig. 3): HEK293T cells expressing IL1RAPL1/IL-1RAcP mutants were co-cultured with cortical neurons and then, accumulation of a presynaptic protein Bassoon in the cortical neurons was analysed (Fig. 3a-d). Because IL1RAPL1-and IL-1RAcP-induced presynaptic differentiation is completely abolished and decreased by B70% in cortical neurons prepared from PTPd-deficient mice, respectively 7,8 , IL1RAPL1/IL-1RAcPinduced presynaptic differentiation is mostly mediated by the mixed, cognate PTPd variants in neurons. On the other hand, to examine postsynapse-inducing activity of PTPd mutants, magnetic beads conjugated with the ECDs of PTPd mutants were co-cultured with cortical neurons, and accumulation of a postsynaptic protein Shank2 around the beads was analysed (Fig. 3e,f). PTPd-stimulating postsynaptic differentiation is mediated by IL1RAPL1, IL-1RAcP and possibly other unidentified postsynaptic organizer(s) in neurons.
Synaptogenic activities of IL1RAPL1 and IL-1RAcP mutants are basically correlated to their affinities to PTPd, which govern stimulation of the receptor activation and signal transmission inside cells. Synaptogenic activities of PTPd mutants are also correlated with their affinities to IL1RAPL1. Remarkably, the R75A mutation in the Ig1 domain of PTPd reduced the synaptogenic activity to B7%. This indicates that the interaction between the IL1RAPL1 Ig3 and PTPd Ig1 domains is critically important for inducing postsynaptic differentiation. In contrast to the affinity of IL1RAPL1, that of IL-1RAcP is not so consistently correlated with synaptogenic activities of PTPd mutants, possibly due to a lack of the interaction between the IL-1RAcP Ig3 and PTPd Ig1 domains. Considering that IL-1RAcP knockout neurons retain nearly full activity of PTPd-induced postsynaptic differentiation, we speculate that the PTPdIL-1RAcP complex might function as a unidirectional organizer that is primarily responsible for inducing presynaptic differentiation in vivo, although IL-1RAcP can promote postsynaptic differentiation of cultured cortical neurons by binding to the meA3-containing PTPd variant, which is specific to IL-1RAcP 7 . The Ig1 domain of PTPd might be universally important for postsynaptic differentiation in vivo, which involves IL1RAPL1 and possibly other unidentified postsynaptic organizer(s).

Discussion
Recently, crystal structures of two different trans-synaptic type-IIa RPTP complexes have been reported 14,25 . One is the complex between Ig1-Ig2 of PTPs and leucine-rich repeat (LRR)-Ig1of TrkC 25 , while the other is that between Ig1-Ig3 of PTPd and the first LRR (LRR1) of Slitrk1 (ref. 14, Supplementary  Fig. 3a). These and our type-IIa RPTP complex structures illustrate that the type-IIa RPTPs can be recognized by structurally distinct architectures. The TrkC-or Slitrk1interacting surface of the RPTPs is separated from the IL1RAPL1-or IL-1RAcP-interacting surface, except that a part of the TrkC-interacting surface is overlapped with the IL1RAPL1interacting surface. The conserved arginine residues in Ig1 of the RPTPs (that is, Arg95 and Arg98 of PTPd and Arg96 and Arg99 of PTPs) are recognized by both IL1RAPL1 and TrkC through similar hydrogen bond interactions (Supplementary Fig. 3b). Despite the separation of these binding surfaces of the RPTPs, pair-wise superposition between any two of these complex structures shows steric clashes between the bound postsynaptic partners, ruling out their simultaneous binding to the type-IIa RPTPs ( Supplementary Fig. 3c). The IL1RAPL1-or IL-1RAcPinteracting residues of PTPd are mostly conserved in the type-IIa RPTPs. Their differences in the specificity to IL1RAPL1 or IL-1RAcP appear to be dependent exclusively on the variation of meA and meB. Our previous analysis of Ptprs and Ptprf cDNA from the developing mouse brain suggested that none of the PTPs variants contains the mini-exon peptide insertion at the meA site 8 and that 90% of LAR lacks the peptide insertion at the meB site 8 . Therefore, IL1RAPL1 and IL-1RAcP induce synaptogenesis specifically through PTPd, at least in the developing brain. On the other hand, the meB insertion is sufficient for binding to Slitrks. They may regulate synaptogenesis mainly through PTPs or PTPd in the developing brain 13,14 , although they can potentially bind to all the type-IIa RPTP variants containing meB. Both the regulation of meA and meB choice by alternative splicing of the type-IIa RPTPs and the mutually exclusive binding among postsynaptic ligands for the type-IIa RPTPs may contribute to sharpening target specificity of central synaptogenesis.
The Ig-Ig interaction between cell-surface receptors is generally important for cell adhesion and communication in immune and neuronal systems. To our knowledge, binding between Ig-containing adhesion receptors is typically mediated by single, homotypic Ig-Ig interactions [26][27][28] in neuronal systems. In contrast, binding between PTPd and IL1RAPL1/IL-1RAcP is mediated by multiple, heterotypic Ig-Ig interactions. The meB insertion of PTPd substantially contributes to these multiple, heterotypic Ig-Ig interactions; meB is located at the junction between the Ig2 and Ig3 domain of PTPd and adjusts their relative spacing and orientation so that they can simultaneously interact with the Ig1 domain of IL-1RAcP or IL1RAPL1 (Fig. 4a). In the IL-1RAcP/IL1RAPL1-bound state, the Ig2 and Ig3 domains of PTPd exhibits a unique conformation that differs from a linear or V-shaped conformation, which is typically observed in a tandem repeat of Ig domains. Furthermore, this unique conformation of PTPd Ig2-Ig3 also differs from the conformation in a Slitrk-bound state 14 . Most PTPd variants expressed in the developing mouse brain at postnatal day 11 contain meB and only 4% of variants lack meB 8 , suggesting that the meB-containing PTPd variants should be mostly utilized for synaptogenesis in the brain. Therefore, the insertion of meB might contribute to allowing PTPd Ig2-Ig3 to form some distinct conformations, depending on the structurally different postsynaptic ligands.
For binding of PTPd to IL1RAPL1, either meA9 or meA6 is essential. Thus, the interface including meA is likely the primary interaction site. Accordingly, the W34A mutation of IL1RAPL1 (at the interface with PTPd Ig2) eliminated binding to PTPd. Moreover, the interaction with meA appears coupled to the third interface involving PTPd Ig1 and IL1RAPL1 Ig3, which is critical for binding between IL1RAPL1 and PTPd and inducing bidirectional synaptic differentiation of cultured cortical neurons, as mentioned above. Comparison between the IL1RAPL1 and IL-1RAcP complexes exhibits a rotational difference in the orientations of the PTPd Ig1-Ig2 unit, relative to the Ig1 domains of IL1RAPL1 and IL-1RAcP (Fig. 4b). Trp34 of IL1RAPL1 is positioned at the pivot for this rotation, which allows PTPd Ig1 to interact with IL1RAPL1 Ig3. In this context, meA also contributes to the multiple, heterotypic Ig-Ig interactions between IL1RAPL1 and PTPd, besides meB, which assists in locating the Ig3 domain of PTPd in the appropriate position for interacting with the Ig1 domain of IL1RAPL1. Consistent with this, the impact of the lack of meB on binding between PTPd and IL1RAPL1 is comparable to that of the PTPd Y273A mutation (deficient in the PTPd Ig3-IL1RAPL1 Ig1 interaction). In contrast, the impact of the lack of meB on binding between PTPd and IL-1RAcP is incompatible to that of the PTPd Y273A mutation, but rather is comparable to that of mutations at the IL-1RAcP Ig1-PTPd Ig2 interface. The W27A mutation of IL-1RAcP, which is equivalent to the W34A mutation of IL1RAPL1, reduced the affinity sevenfold but did not eliminate the binding. Therefore, for binding to IL-1RAcP, the PTPd Ig3-IL-1RAcP Ig1 interface is likely the primary interaction site, and the meB insertion may serve as the adjustable linker to locate the Ig2 domain of PTPd in the appropriate position for interacting with the Ig1 domain of IL-1RAcP (Fig. 4a). IL-1RAcP functions in both immune and neuronal systems, whereas IL1RAPL1 is exclusively expressed in neuronal systems, suggesting that IL1RAPL1 has evolved to be more specialized for neuronal systems than IL-1RAcP. Correspondingly, IL-1RAcP binds to PTPd weaker than IL1RAPL1 but has a broader specificity to the meA variants. In contrast, IL1RAPL1 has more strict specificity to meA and can bind to PTPd more tightly than IL-1RAcP. The affinity difference generated by the splice-insert signaling codes is substantially high for binding between PTPd and IL1RAPL1 but modest for binding between PTPd and IL-1RAcP. On the other hand, in our synaptogenic assays, the strength of the induced synaptic differentiation seems non-linearly correlated with the affinity; rather, there seems to be the threshold of the affinity for inducing the synaptic differentiation, which might contribute to controlling appropriate synaptic target selection to avoid misconnections to target cells with low affinity ligands. Further functional studies including identification and characterization of additional regulatory factors are awaited to discuss the activation and signaling mechanisms of IL1RAPL1/IL-1RAcP and PTPd for inducing synaptic differentiation.  crystallization experiments except for PTPd Fn1-Fn2 were performed at 20°C. Temperature shift from 4 to 28°C drastically improved the crystal size of PTPd Fn1-Fn2. Crystals were flash-frozen in liquid N 2 , followed by soaking in the reservoir solutions supplemented with the following cryo-protectants: 25% glycerol for PTPdA3B À Ig1-Fn2 and the PTPd Ig1-Ig2IL1RAPL1-ECD complex; 25% ethylene glycol for PTPd Fn1-Fn2; 27% ethylene glycol for IL1RAPL1; 35% xylitol for the PTPd Ig1-Fn2IL-1RAcP-ECD complex; 20% ethylene glycol for the PTPd-ECDIL1RAPL1-ECD complex.
Crystallography. All data were collected at 100 K at BL41XU in SPring-8, and processed with HKL2000 (ref. 30) and CCP4 program suite 31 . Data collection and refinement statistics were summarized in Table 1. Resolutions were estimated, basically based on I/sI values (B2). To improve the quality of the atomic models, the resolutions of PTPd-ECDIL1RAPL1-ECD and PTPdA3B À Ig1-Fn2 were set to be as high as possible. The resolution of PTPd Fn1-Fn2 was limited by the size of the detector. We first determined a crystal structure of the PTPd Ig1-Ig2IL1RAPL1-ECD complex by the molecular replacement method using IL-1RAcP-ECD (PDB:4DEP, 3O4O) 32 SPR analysis. SPR equilibrium experiments were carried out by using Biacore T200 (GE Healthcare) at 25°C in 10 mM Hepes (pH 7.9) containing 150 mM NaCl, 0.05% Tween-20. IL1RAPL1-ECD-His 6 was immobilized on a CM5 sensor chip by the amine-coupling method at a density of about 2,000 resonance units (RU), whereas IL-1RAcP-ECD-Fc was captured at a density of about 700 RU by anti-Fc antibodies that were immobilized on a CM5 sensor chip. Splicing variants or mutants of PTPd Ig1-Fn1 were injected as the analytes with the concentration range from 15.6 to 4,000 nM. The IL1RAPL1-immobilized sensor chip was regenerated by 10 mM NaOH. The sensorgrams are shown in Supplementary Fig. 4.
Cell cultures and co-culture assay. A central nervous system-restricted isoform of IL-1RAcP, IL-1RAcPb, was used for HEK293T cell-based analyses of IL-1RAcP mutants. Primary cortical cultures were prepared from mice at E18 (ref. 38). The cortical cells were placed on coverslips coated with 30 mg ml À 1 poly-L-lysine and 10 mg ml À 1 mouse laminin at the density of 5 Â 10 5 cells per well for co-culture assay. The cells were cultured in Neurobasal-A supplemented with 2% B-27 supplement (Invitrogen), 5% FCS, 100 U ml À 1 penicillin, 100 mg ml À 1 streptomycin and 0.5 mM L-glutamine for 24 h, and then cultured in the same medium without FCS. Cultures of HEK293T cells were maintained in DMEM supplemented with 10% FCS 38 . Expression vectors for mutated forms of PTPd-ECD-Fc, FLAG-tagged IL-1RAcPb and FLAG-tagged IL1RAPL1 were generated by PCR-based mutagenesis using pEB6-PTPd-ECD-Fc, pFLAG-IL-1RAcPb and pFLAG-IL1RAPL1 (ref. 7) as templates, respectively. Expression vectors for FLAGtagged IL1RAPL1 and IL-1RAcP mutants were transfected to HEK293T cells using NanoJuice transfection reagent (Novagen). After 2 days of culture, the transfected cells were washed with PBS containing 2 mM EDTA, and incubated with the same buffer at 37°C for 10 min. Dispersed cells were plated onto cortical neurons at days in vitro (DIV) 8. After 24 h of co-culture, cells were fixed for immunostaining with mouse anti-Bassoon (Stressgen, 1:300) and rabbit anti-FLAG (Sigma, 1:1,000) antibodies. Fc and ECDs of PTPd mutant proteins fused to Fc in HEK293T cell culture medium were bound to Protein A-conjugated magnetic particles (smooth surface, 4.0-4.5 mm diameter; Spherotech). Beads coupled with Fc or Fc fusion proteins were added to cortical neurons (DIV 14). After 24 h, cultures were fixed for immunostaining with rabbit anti-Shank2 antibody (Frontier Institute, 1:200).
Image acquisition and quantification. Images of fibroblast-neuron or beadneuron co-culture experiments were acquired with a confocal laser-scanning microscope TCS SP5II (Leica) under constant conditions as to laser power, iris, gain, z-steps and zoom setting throughout the experiments. The images were collected from at least two separate cell cultures. All quantitative measurements were performed with ImageJ 1.36b software 39 . The intensities of immunostaining signals for Bassoon or Shank2 were measured as the optical mean density within circles of 30-and 7-mm diameters enclosing transfected HEK293T cells and coated-beads, respectively. Statistical significance was evaluated by one-way analysis of variance followed by post hoc Tukey's test.